Skip to content

Pooling four PLFS years: does it change anything?

The CY2025 engineering result rests on a ₹6 lakh cell of 34 people. Pooling more years is the obvious way to buy precision, so it was done separately, end to end, and compared. Run pooled_engineering.py.

Short answer: it confirms CY2025, buys a 26% tighter interval, and nudges the estimate up. No finding changes. One assumption I nearly got wrong is worth more than the extra precision.


Release Used Why
calendar_2022calendar_2025 Four consecutive, non-overlapping calendar years
All annual_* They overlap the calendar releases. annual_2022_23 covers July 2022 to June 2023, spanning two of them. Using both series counts the same survey periods twice.
calendar_2021 Carries no tedu_lvl, pas or ern_reg at all — zero engineering rows

Pooling buys less than four years suggests. calendar_2025 holds 1.15m persons against roughly 420k in each earlier release — PLFS expanded its sample — so 2025 alone carries about 45% of the pooled rows. The 22–24 engineering window goes from 1,042 to 2,338, not to 4,000-odd.


The deflator, and the mistake it nearly caused

Section titled “The deflator, and the mistake it nearly caused”

Wages are nominal, and a fixed ₹50,000 line is a higher real bar the further back you look. Rather than import an external price index this data cannot verify, earlier years are restated using an internal one: the wage of all regular salaried workers, any education, ages 15–59.

Release n Median Mean p75
calendar_2022 42,163 ₹14,100 ₹20,525 ₹27,000
calendar_2023 43,569 ₹15,000 ₹21,683 ₹28,350
calendar_2024 44,726 ₹15,000 ₹22,529 ₹28,950
calendar_2025 118,277 ₹15,000 ₹23,497 ₹29,500

The median is unusable as a deflator. It sits at exactly ₹15,000 in three consecutive years — that is the round-number heaping already documented in this analysis, not stable pay. Deflating on it would have silently assumed away all nominal wage growth and produced a confident, wrong conclusion that “median nominal pay for salaried India did not move.”

The mean rises 14.5% over three years, about 4.6% a year; p75 rises 9.3%. The mean is used. On that basis, ₹50,000 in 2025 is worth about ₹43,700 in 2022 rupees.


Basis n FSUs In formal work Formal and ≥₹6L 95% CI
CY2025 only 1,042 916 34.65% 4.42% [2.19, 7.19]
Pooled 2022–25, nominal 2,338 2,085 35.96% 5.14% [3.41, 7.14]
Pooled 2022–25, deflated 2,338 2,085 35.96% 5.59% [3.82, 7.61]

Year by year, which is what the pooled figure averages over:

Release n In formal work ≥₹6L nominal ≥₹6L deflated Cell n
calendar_2022 447 38.68% 2.37% 2.97% 13
calendar_2023 436 32.26% 5.39% 6.12% 18
calendar_2024 413 37.92% 7.98% 8.41% 26
calendar_2025 1,042 34.65% 4.42% 4.42% 34

Formal employment is stable — 32–39% across four years, with no trend. That figure is solid.

The ≥₹6L series is noise, not trend. 2.97 → 6.12 → 8.41 → 4.42 does not describe anything; those cells hold 13, 18, 26 and 34 people. Deflation lifts the early years slightly, as it should, but does not straighten them. Nothing here supports a claim about engineering wages improving or worsening over 2022–25.

Wage by tenure, formal workers aged 21–34

Section titled “Wage by tenure, formal workers aged 21–34”
Tenure Basis n Median ≥₹50k
Under 1 year CY2025 182 ₹25,000 11.19%
Pooled 408 ₹26,500 14.59%
1–3 years CY2025 762 ₹35,000 24.38%
Pooled 1,728 ₹35,000 20.55%
Over 3 years CY2025 1,216 ₹45,000 42.85%
Pooled 2,547 ₹45,000 41.84%

These medians are weight-replicated and so sit above the unweighted APPROX_QUANTILES figures published elsewhere in this analysis — ₹45,000 here against ₹42,000 in README.md for the same cell. Neither is wrong; they are different estimators, and the corpus has not standardised on one. See ../ASSUMPTIONS.md D4.

On a like-for-like basis the medians are near-identical. The progression shape — roughly 1.7× from entry to 3+ years — holds in both.


n ≥₹6L cell Interval width
CY2025 only 1,042 34 5.01pp
Pooled 2022–25 2,338 91 3.73pp

26% tighter, and the point estimate moves 4.42% → 5.14% nominal, 5.59% deflated. Every version sits inside every other version’s interval, so the two analyses agree.

The cost is that the pooled figure is no longer current. It describes an average engineering graduate over 2022–25, and mixes a labour market before and after whatever changed. Given that the year-to-year variation is noise, that cost is small — but it is a real one, and it is why CY2025 remains the headline with this as the robustness check rather than the other way round.


The engineering number feeds a separate analysis, where NIRF’s ranked colleges claim 62,613 graduates earning ≥₹6 lakh. Applying each estimate to AISHE’s 784,517 engineering graduates:

Basis Rate National pool NIRF’s claim as a multiple
CY2025 only (current headline) 4.42% 34,676 1.81×
Pooled, nominal 5.14% 40,324 1.55×
Pooled, deflated 5.59% 43,855 1.43×
Pooled deflated, CI low 3.82% 29,969 2.09×
Pooled deflated, CI high 7.61% 59,702 1.05×

The over-claim survives across the whole range, but its size moves a lot — from 1.05× at the optimistic end to 2.09× at the pessimistic one. The qualitative finding, that NIRF’s ranked engineering colleges claim more ₹6 lakh fresher jobs than the country appears to contain, holds at every point including the upper confidence bound. The specific multiple does not, and should be quoted as a range rather than a number.

That is a stronger position than the single-year analysis alone supported, and it is the main thing pooling bought.


  • The formal employment rate, 34.6% vs 35.96%. Stable across all four years.
  • The wage progression shape, roughly 1.7× from entry to 3+ years. (This line said 1.9× until the tenure age-bound fix; 1.68× is the corrected figure.)
  • The entry median, ₹25,000 vs ₹26,500.
  • The conclusion that the ≥₹6L cell is the binding constraint. Pooling takes it from 34 to 91, which helps, but it is still the smallest number in the analysis and still governs what can honestly be claimed.